STL10 fails to load unlabeled data

1

Hello,

I am trying to train a model on the unlabeled split of the STL10 dataset. When running the following code I run into an error:

stl_transforms = transforms.Compose([transforms.ToTensor(),
                                    transforms.ToPILImage(),
                                    transforms.ToTensor()])

unsupervised_data = datasets.STL10(root = '../data', split =  'unlabeled', transform=stl_transforms, download=True)

unsupervised_train_loader = torch.utils.data.DataLoader(unsupervised_data,
                             batch_size=args.batch_size, shuffle=True)

for batch_idx, (data, _) in enumerate(unsupervised_train_loader):

In this case I get: TypeError: batch must contain tensors, numbers, dicts or lists; found <class ‘NoneType’>

Interestingly, if I set split to ‘train’ or ‘train+unlabeled’ , everything works well.

Any help would be appreciated!

2

Hey @Loerbri
I have tried the code snippet out. I do not seem to have a problem with the working.

Would you mind sharing a colab notebook where the issue is reproducible?